I'm an Applied Mathematics, Computer Science & Economics student at Yale. I build systems that turn market questions into reproducible evidence—from modern C++ pricing engines and leakage-aware backtests to commodity and prediction-market research packages.
My current interests include volatility surfaces, market microstructure, robust portfolio risk, point-in-time data, and the engineering that makes quantitative results auditable.
| Project | Current completed evidence | Boundary |
|---|---|---|
| quant-pricer-cpp v0.4 | Portfolio risk at 20.18× / 20.25M positions/s; exact stress at 27.92× / 32.13M cells/s; worst QuantLib price error 3.91e-14; 32/32 concurrent replays bitwise identical |
Deterministic Black–Scholes European risk/stress; recorded Apple M3 Pro protocol—not forecasting or live P&L |
| microalpha v0.3 | 2017–2025 OOS risk case: Sharpe 1.04, volatility 11.37%, max drawdown −15.31% vs market 0.83, 19.25%, −34.22%; Audit Lab catches four injected backtest failures |
Reproducible risk-management and correctness evidence; no differential-return claim (p=0.467) |
| Project | What it demonstrates |
|---|---|
| quant-pricer-cpp | C++20/Python derivatives system spanning portfolio risk and exact stress, analytic pricing, Monte Carlo/QMC, PDE, Heston, exotics, Greeks, and independent numerical validation |
| microalpha | Event-driven quantitative engineering lab pairing a real-data market-risk case with point-in-time guards, t+1 execution, costed walk-forward evaluation, and an adversarial Audit Lab |
| Wheat Tightness | Official-source commodity research system with an explainable scorecard, scenarios, deck, provenance, and frozen as-of conclusions |
| Magnetic Corne Cases | Original 3D-printable keyboard, travel-case, and wrist-rest system used by the split-keyboard community |
- Chronology before performance — point-in-time inputs, explicit clocks, and no silent lookahead.
- Costs and uncertainty are first-class — turnover, slippage, borrow, confidence intervals, baselines, and stress tests belong beside the result.
- Independent checks beat confident prose — cross-engine parity, walk-forward evaluation, falsification tests, and reproducible artifacts.
- Claims match the evidence — descriptive risk reduction, synthetic correctness controls, and investable alpha are kept explicitly distinct.
C++20 · Python · SQL · kdb+/q · CMake · OpenMP · NumPy/pandas ·
PyTorch · CVXPY · WRDS/CRSP · GitHub Actions